MCP is awesome but also tricky. Right now if anything goes wrong I can get the log of back and forth between my client and the pebble cloud agent, but i can’t figure out how to see the agent’s reasoning, what tool it tried to call, or what failure actually occurred. Was it an authentication issue? Service failure? Malformed request? Dropped session?
Is there a way to dig into further debugging detail than the “trace timeline” and the ring sync inspector’s GenericFailure?
If you tap the share icon in the new bug report screen, you can download logs from the Pebble app. It’s not perfect right now. We’ll definitely add better MCP analysis but that might help in the short term!
Thanks fpr the tip @eric!
Unrortunately the zipped logs don’t clarify what tool was called, why, or what the error was. I’d submit a pr but those parts are server side…
See Validate mcp server name will survive composite tool name round-trip by ohthehugemanatee · Pull Request #367 · coredevices/mobileapp · GitHub for the issue and my patch to resolve it. In the meantime, users beware: if your MCP server “name” has spaces or special characters, the agent receives the wrong MCP name to call and it fails. Stick with alphanumeric and “-” and it’ll work.
… and I just put 2 and 2 together about which @eric answered my post.
You’re doing fantastic work, thank you for all of it.
The best part of the Index experience so far has been the sense that this device is so open because your team really cares about users, you understand that flexibility surfaces the best use cases, and genuinely welcome that feedback/contribution. It’s refreshing in the age of narrative, and I know it doesn’t happen by accident. Thank you!
Thanks so much for the PR. We’ll check it out. I appreciate your note as well. Remember that today’s day one. It’s just going to get better!